A Faster Algorithm for Finding Negative Cycles in Simple Temporal Networks with Uncertainty

Temporal constraint networks are data structures for representing and reasoning about time (e.g., temporal constraints among actions in a plan). Finding and computing negative cycles in temporal networks is important for planning and scheduling applications since it is the first step toward resolving inconsistent networks. For Simple Temporal Networks (STNs), the problem reduces to finding simple negative cycles (i.e., no repeat nodes), resulting in numerous efficient algorithms. For Simple Temporal Networks with Uncertainty (STNUs), which accommodate actions with uncertain durations, the situation is more complex because the characteristic of a non-dynamically controllable (non-DC) network is a so-called semi-reducible negative (SRN) cycle, which can have repeat edges and, in the worst case, an exponential number of occurrences of such edges. Algorithms for computing SRN cycles in non-DC STNUs that have been presented so far are based on older, less efficient DC-checking algorithms. In addition, the issue of repeated edges has either been ignored or given scant attention. This paper presents a new, faster algorithm for identifying SRN cycles in non-DC STNUs. Its worst-case time complexity is O(mn + k²n + knlog n), where n is the number of timepoints, m is the number of constraints, and k is the number of actions with uncertain durations. This complexity is the same as that of the fastest DC-checking algorithm for STNUs. It avoids an exponential blow-up by efficiently dealing with repeated structures and outputting a compact representation of the SRN cycle it finds. The space required to compactly store accumulated path information while avoiding redundant storage of repeated edges is O(mk + k²n). An empirical evaluation demonstrates the effectiveness of the new algorithm on an existing benchmark.